A Machine Learning Approach Using Classifier Cascades for Optimal Routing in Opportunistic Internet of Things Networks

Abstract

Routing in Opportunistic Internet of Things Network (OppIoT) is an involved problem, because the network is intermittently connected and source to destination end-to-end paths are non-existent. Moreover, Machine Learning (ML) has recently achieved great success in multiple domains and is now being applied to automate routing in Opportunistic Networks (OppNets) which are similar in characteristics to OppIoT, through protocols such as MLProph and KNNR. In this paper, we utilize cascade learning, a form of ensemble based ML, for improved routing in OppIoT. Through simulations we show that our proposed protocol called Cascaded Machine Learning based routing protocol (CAML), outperforms existing ML based protocols (MLProph and KNNR), and traditional well-performing protocols (HBPR and PRoPHET), on a wide range of performance metrics including message delivery probability, average hop count, packets dropped and network overhead ratio.